Opinion Polarity Classification

Polarity classification aims to determine the orientation of the opinions, which is a nature extension of opinion retrieval. The task was introduced at TREC 2007 Blog Track. Its aim is the classification of the opinionated relevant posts into positive, negative or mixed class according to the polarities of opinions within blog posts.

Proposed and implemented an opinion polarity classification system. The positive and negative query-relevant opinions are identified respectively. The strengths of polarized opinions are evaluated to determine whether an query-relevant opinionated document is positive, negative or mixed (when both positive and negative opinions are equally strong).

Witnessed that negation plays a central role in opinion polarity classification and thus proposed a novel algorithm to calculate the scope of a negation term within the context of sentences. The state-of-the-art negation-handling methods simply assume the scope of a negation term t is either several terms following t or the whole clause (or even sentence) containing t. However, the identification of the scope of negation is complicated and challenging. For example, a negation term may not affect its following words close to it but affect the following words far from it. The algorithm significantly outperformed the state-of-the-art methods in determining the scope of negation and the utilization of our algorithm enhanced the sentiment analysis performance.

The basic ideas are based on the following papers.